- Claude 4 includes two flagship models: Opus 4, geared toward complex tasks, and Sonnet 4, more efficient for everyday use.
- Both excel in coding, advanced reasoning, and memory management, outperforming competitors in benchmarks like SWE-bench.
- The new models incorporate extended thinking, parallel tooling, and improved developer APIs.
- They feature enhanced security measures, flexible usage options, and a pricing structure based on capacity.
The launch of Claudia 4 Anthropic's innovation has generated significant interest in the artificial intelligence sector. At a time when the race for technological leadership is accelerating every day, the American company is reinforcing its commitment with two improved models that seek to optimize the programming, advanced reasoning, and operation of autonomous agents.
These new features consolidate Anthropic as one of the great leaders in AI, betting heavily on functionalities that range from greater coding capacity to more advanced memory management, geared toward both technology companies and demanding individual users and developers.
New generation of models: Opus 4 and Sonnet 4
Anthropic's latest family of models comes in the form of two great proposals: Claude Opus 4, designed for highly complex and prolonged tasks, and Claude Sonnet 4, which follows the line of solid results but with greater efficiency for everyday cases.
What distinguishes this model is its ability to work autonomously for hours without losing precision, according to the experiences of companies like Rakuten, which reported code refactoring sessions lasting up to seven hours at a time with high performance. Memory management allows for the storage of relevant information and maintenance of context even in projects with thousands of steps, making Opus 4 an ideal foundation for intelligent agents that require sustained autonomy.
On the other hand, Claude Sonnet 4 It focuses on everyday tasks, maintaining a balance between power and practicality. With a score of 72,7% in SWE-bench, this version significantly improves on its predecessor (Sonnet 3.7), excelling in efficient coding, instruction understanding, and performance stability, even for less experienced users.
Innovations in reasoning and memory
One of the functions that has generated the most interest is the extended mode of thinking with the use of tools, which allows models to switch between internal reasoning and the use of external utilities, such as web searches, in a fully integrated way. This results in a improved troubleshooting, where reflection and the practical action of executing code or consulting additional information are combined.
Both models can use multiple tools simultaneously and access local files to manage contextual memory during extended sessions. This feature improves performance in complex, ongoing projects, helping developers maintain a working thread without losing important data.
Another notable innovation is the memory optimization. Claude 4, when you have access to local files, you can create and maintain “memory files” that allow you to track the progress of extensive tasks, store tacit learning, and ensure consistency in long-term work.
Developer tools and improved API
Along with the new models, Anthropic has officially presented Claude Code, a command-line tool that integrates AI directly into the developer environment. This application is compatible with popular IDEs such as VS Code and JetBrains, allowing for real-time editing suggestions, automated reviews, and task management from within the editor.
Claude Code now offers support for running background tasks using GitHubActions, and its extensible SDK enables developers to build their own intelligent agents on top of Claude's core. All of this is supported by a revamped API that includes code execution, connectivity via MCP, a file API, and caching functions that optimize the use of prompts for up to an hour.
Performance, safety and prices
Claude 4 offers response modalities that include Instant answers for quick queries and extended thinking for complex analysisThis flexibility allows the model to be adapted to different types of tasks, from the simplest to long-running workflows and large volumes of data.
In terms of to maximise security and your enjoyment., the firm has reinforced protection protocols, implementing standards such as AI Safety Level 3 (ASL-3) in Opus 4 and additional layers to reduce risks in everyday use. The goal is to minimize unwanted behaviors, shortcuts, or errors, and prevent problematic uses such as the generation of dangerous content or improper data manipulation.
Prices remain in line with previous versions: Opus 4 It costs $15 per million input tokens and $75 per million output tokens, while sun 4 priced at $3 and $15 respectively. Both models are available through the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and various subscription plans for businesses and individuals.
This advancement in AI technology positions Claude 4 as a very solid option for managing software projects, solving complex problems, and collaborating autonomously with humans, thanks to its focus on memory, security, and extended thinking.